AI Guides › Step-by-step guides
By Nigel Guy · 7 min read
Most people who post across four or five platforms judge what is working by memory: the post that got a nice comment, the video a friend mentioned. Each platform's own analytics sits in a different app with different definitions, so nobody compares them, and the week gets planned on gut feel. Connecting Claude to Metricool fixes the data problem, but only if you also stop Claude publishing anything you have not seen.
The rule: let Claude read everything and schedule nothing without showing you the exact post, platform and time first.
| You need | Detail at time of writing |
|---|---|
| A Metricool account | Free plan works with the MCP connector. Free covers 1 brand, 20 posts a month, 30 days of analytics and 5 competitor profiles, and does not connect LinkedIn or X. |
| A paid Metricool plan, if you need more | Starter adds LinkedIn, unlimited analytics history, best-time-to-post and ads analytics. Metricool lists prices in euros and dollars, not pounds: Starter is shown from €16–€29 a month (roughly £14–£25), Advanced from €43–€130. X is a separate add-on (€10 a month per account). Check the pricing page before you buy. |
| A Claude account | Free, Pro, Max, Team or Enterprise all support remote connectors. Free allows only one custom connector at a time. Claude Code needs Pro or Max. Anthropic bills Pro in US dollars ($20 a month plus VAT), so the sterling amount moves with the exchange rate. |
| Admin rights, on Team or Enterprise | Only Owners and Primary Owners can add connectors on a Team plan. |
Do this in Metricool, not Claude. Create your brand, then connect Instagram, TikTok, LinkedIn, Facebook, YouTube and whatever else you use. For ads data, connect your Meta, Google and TikTok ad accounts too. Through the connector, Claude can only read ad results. It cannot change campaigns.
If you want competitor comparisons later, add the competitor accounts in Metricool now. It only gathers their public data after you add them, so a competitor added today has no history yet.
Claude.ai or Claude Desktop (one connection covers both):
If Metricool is not in your directory, add it as a custom connector with the URL https://ai.metricool.com/mcp.
Claude Code:
claude mcp add --transport http metricool https://ai.metricool.com/mcp
Then start claude, run /mcp, choose Metricool, select Authenticate, and grant access in the browser.
Neither route needs an API key; sign-in is OAuth.
In a new chat, click +, open Connectors and switch Metricool on. While you are there, check the tool permissions. Leave the planner tools on "ask before use" rather than always-allow, so every scheduling action waits for your click.
Start by checking what Claude can actually see. A missing account or a 30-day history cap changes how far you can trust every later answer.
Nothing to fill in:
You are my social media analyst with read access to my Metricool account.
Goal: confirm what data you can see before we analyse anything.
1. List every brand and every connected social profile and ad account you can access.
2. For each, give the earliest and latest date you hold data for.
3. Flag any profile with fewer than 30 days of data, or with gaps.
4. Only then compare engagement rate across my platforms for the last 30 days. Name the strongest and weakest platform, and say how Metricool defines engagement rate for each if the data shows it.
Output: a table (profile | platform | data from | data to | notes), then three sentences on strongest and weakest.
If a platform you would expect is missing, say so; do not assume it has no activity.
Before answering, check that every figure came from a tool result, not an estimate.
This is the weekly habit. Fill in [DAYS] (normally 7) and [PLATFORMS], or write "all".
Act as a blunt content analyst. Use only my Metricool data for the last [DAYS] days on [PLATFORMS].
Produce, in this order:
1. Top five posts by engagement across platforms: platform, date and time posted, format, opening line.
2. What those five share: subject, format, length, posting time. Only call something a pattern if at least three of the five show it.
3. The three weakest posts, with the likeliest reason each underperformed.
4. Best posting windows per platform, taken from Metricool's best-time data, not your own guess.
Format: one page, headed sections, a table for item 1.
Label every inference "[guess]" and every figure "[data]".
If [DAYS] or [PLATFORMS] is missing, ask me before running anything.
Self-check before replying: are any posts counted twice across platforms? Is any "pattern" built on fewer than three posts?
Fill in [TOPIC] and [NUMBER] (ten is plenty).
You are a content strategist working from evidence, not trends.
Using my Metricool data from the last 14 days, find my best-performing posts about [TOPIC]. If fewer than three exist, tell me and stop.
Then propose [NUMBER] new post ideas built on what those posts did well. For each: hook (one line), format (carousel, short video, text post, etc.), best-suited platform, and which existing post it draws on.
Return a numbered list. Do not suggest formats I have never posted unless you mark them "untested".
Check before answering: does every idea trace back to a named post?
Fill in [CAPTION], [MEDIA], [PLATFORMS] and [DAY].
You can create scheduled posts in my Metricool planner. Treat this as a two-stage job.
Stage 1, draft only:
- Look up my best posting time on each of [PLATFORMS] for [DAY].
- Show a table: platform | date | local time (UK) | caption as it will appear | media attached.
- Point out anything a platform may reject: caption length, missing video, wrong aspect ratio.
- Then stop and wait.
Stage 2 runs only after I reply "yes, schedule". Create the posts exactly as shown, then read them back from the planner to confirm.
CAPTION: [CAPTION]
MEDIA: [MEDIA]
If the media is not attached or uploaded to Metricool, ask for it. Do not schedule a text-only version instead.
Fill in [COMPETITOR_1] and [COMPETITOR_2], which must already be added in Metricool (Step 1).
Using Metricool competitor data for [COMPETITOR_1] and [COMPETITOR_2] over the last 30 days, alongside my own:
1. Which of their posts beat their own average, and the likely reason.
2. Formats they use that I do not.
3. One specific test I could run next week, with how I would judge it.
Mark anything not drawn from the data as "[guess]". If either competitor has under 30 days of data, say so first.